When do we (or should we) use HSV color space and when do we use CIE-Lab color space?
A perceptual uniform color space ensures that the difference between two colors (as perceived by the human eye). It is proportional to the Euclidian distance within the given color space.
You may think that the lightness component of the HSL color system or the value component of HSV will solve this problem, It's wrong.
The L component of the HSL and the V component of HSV describe the brightness of a color relative to a base color. If this base color is blue you will get a darker color than if this base color is yellow, HSL and HSV are very useful if you need to create a lighter or darker version of a color but aren’t very useful if you want to know how bright a color is.
But, the relative perceptual differences between any two colors in Lab* can be approximated by treating each color as a point in a three-dimensional space (with three components: L*, a*, b*) and taking the Euclidean distance between them. Lab was created to satisfy the perceptual uniformity property.
The history of HSV/HSL was that it had a low CPU overhead and was a 'good enough' approximation at the time. Essentially it was the stop gap used in the industry until there was enough CPU power to enable a Lab workflow.
So the answer is Lab is the way to go. Though there are plenty of ways to mimic some of the Lab behaviours using the blending modes in RGB. But definitely of HSV/Lab then it should be Lab every time.
Both have their usage:
HSV / HSL colour models offer a quite useful separation of hue and saturation which is really important for creative operations such as colour correction and colour pickling for instance (as a matter of fact most colour pickers offer a way of adjusting / defining your colour through a HSV colour model). However the way those models compute Brightness / Lightness is incorrect and yields non uniformity: http://www.poynton.com/notes/colour_and_gamma/GammaFAQ.html#HSI
CIE Lab is a perceptually uniform colourspace where the Lightness Y is properly computed thus will result in accurate colour transformations if you perform them in this colourspace which is important in colour science, and probably less in creative applications where it is almost impossible to do anything with the opponent a and b axis.