LPI filter framework redesign

Hi all, I just revisited the LPI filter framework for the first time in many years, and...err...it's pretty bad :) I wrote this originally to simulate the effect of a specific point-spread-function on an image--and for that purpose it works OK. But for restoring images based on a kernel, the interface doesn't work. I'd like to redesign this functionality, and would like to have your input. Does anyone here have much experience in this type of filtering and deblurring images? My bigger aim is to implement fairly robust blind deconvolution. Stéfan