I suggest you check out sci.crypt, but to quote the relevant bit of discouragement from the sci.crpyt FAQ:

  Among professionals, a common rule of thumb is that if you want to 
  design a cryptosystem, you have to have experience as a cryptanalyst
+.
[download]

This advice applies as much to modifying existing cryptosystems as it does to creating new ones. Unless you have a serious theoretical background in probability and number theory, as well as in cryptanalysis, don't bother. In this problem domain, our resident monks will be about as helpful as the well-intentioned folks from rec.glass-blowing.

   MeowChow                                   
               s aamecha.s a..a\u$&owag.print