The only available tool able to create a RAW file is, as far as I know, a camera, and there are no standard tools to manipulate it. I've heard that's why RAW files are used in different photography competitions to prove a photographer has not manipulated a submitted photograph.
However, RAW files are just data files and any bits can be manipulated.
What makes RAW files difficult to be an object of manipulation?
Nothing makes raw files difficult to manipulate for someone with the right expertise and tools. It's just that there aren't many folks around who have those tools and expertise.
The tools needed to manipulate a raw file into a jpeg are much more widespread and well known than those needed to manipulate a raw file into a different raw file. That is probably where the perception that raw files are more difficult to manipulate comes from: most organizers of such contests are more familiar themselves with how to produce heavily manipulated jpegs from raw files. Most of them are probably not aware that raw data can be manipulated at all, much less how one would go about doing it. I mean, they don't even understand what 300 dpi (doesn't) mean in a display-size agnostic digital environment.
Ironically, the news organization Reuters has it the other way around: They will only accept images that (appear to) have been generated as jpegs in camera at the time the images were shot.
A RAW file is little more than a container for the output of a camera sensor. It has to be processed into an image which gives it full color information at each pixel. As such there are no programs intended to manipulate a RAW file since it is meant as input to RAW conversion software.
Since it is just made of bits like any other digital file one can of course change any portion with a binary editor for example. What is harder would be to make coherent changes to the RAW file which that the changes would appear natural. You cannot crop a RAW file because the image size would no longer produce what the camera does. Adding or removing objects from the scene would require one to do the inverse transformation of a RAW conversion specific to a particular camera.
There are better formats for lossless, high bit-depth image storage and exchange. The main benefit of raw files is that they contain minimally-processed sensor data. So there's no compelling reason for anyone to put the (fairly significant) effort in to writing the code to write raw files. This means that submitting the raw files is a practical way to demonstrate (i) possession of the original shot and (ii) what the original shot looked like. It's poor evidence of legal ownership of the original.
RAW files are hard to manipulate because there are no tools for this.
There are no tools because manipulating them is pointless.
RAW files don't hold standard images. They hold the data read directly from the sensor of one specific camera model. They need to be processed in a way specific to each camera model to get a standard image.
In order to "display" a raw file, you must have specific details of the camera model it came from—this is why RAW converters like Lightroom need to get an update for each new released camera. In contrast, JPEGs or PNGs are designed to be displayable on any device without needing to know where they came from. They are meant to be able to hold any image.
I hope this makes it clear that there is absolutely no point in producing RAW files in any other way than directly in the camera. (Unless you want to commit fraud, or unless you want to reverse engineer a specific RAW format to understand a camera better, or to produce your own RAW converter for it.)
A typical camera sensor does not capture RGB pixels, but instead captures input distinct red-sensing pixels, green-sensing pixels, and blue-sensing pixels at slightly different locations; a raw file will report the values of the individual pixels as captured.
When a raw file is converted to an RGB pixel format, each pixel in the output file will typically be a weighted and filtered average of a number of pixels on the original sensor. Once the data is converted, each pixel in the resulting file will be capable of independently representing any color. If one wants to e.g. adjust the saturation in a file of RGB pixels, each individual pixel's red value can be based upon its blue and green values, and likewise adjust each pixel's blue based upon its red and green, and adjust its green based upon its red and blue.
If one wanted to apply a white-balance adjustment to a raw file, however, one wouldn't be able to adjust the color of individual pixels, since each individual pixel is only capable of sensing a single lightness value. If one wants to reduce the saturation of a raw image of a red object, it wouldn't be possible to increase the blue and green values of all the red-sensing pixels; instead, one would have to increase the reported values for blue-sensing and green-sensing pixels that were near brightly-lit red-sensing pixels. Such operations are not difficult, but each time they are applied will degrade the image a little more. By contrast, the act of converting sensor data to an RGB picture is generally lossy, but such loss need only be incurred once.
Because there is no need to do it.
Advanced manipulation programs usually are non destructive, so programs usually use original image, but save manipulation in an additional file (or database). For simple editing, the sidecar .xmp is used.
There is huge advantage to have non destructive workflow. One for all: easier to backup, but you can always have the original information to do further manipulations, without losing anything.
I think UFRaw can also save raw.
Technically it is not difficult to create them: common raw files are just compressed TIFF files, with some well know interpretation of color and geometry of pixels (because it is used by the readers), and with some additional EXIF information.
