Yes, if you need an approximate pdf. I've found that when I'm working non-parametrically (or in robust statistics) I like to stay in cdf space or use quantile f

Yes, if you need an approximate pdf. I've found that when I'm working non-parametrically (or in robust statistics) I like to stay in cdf space or use quantile functions more than trying to use those nasty derivatives.