What you're describing is 'deconvolution' and there exists algorithms designed to do exactly this (see for example, Richardson-Lucy deconvolution). However, you need to either know or make some assumptions about the 'blur' for it to work.

There is software that Canon releases for high-end cameras and lenses that does something similar. Becuase they know exactly the type of blur that their lenses create at different points on the frame for different focal distances, they use this information to remove some of that lens blur from the digital image. Canon call this 'digital lens optimizer'. See here and here for more info.