You could start with the AAIndex database which is a collection of a variety of different amino acid propensities. You then have to decide which one to take and represent each of your fragment as a propensity vector. As mentioned before, you need some similarity measure between two such vectors. An obvious one would be
s(v1,v2)=1-|corr(v1,v2)| , corr being the pearsson correlation coefficient. For clustering, I would start with linkage clustering. R has some nice implementations for this.
Have a look at the EMBOSS suite of tools (i.e. embedded in usegalaxy.org). It at least gives you the properties either in signatures or in raw values.....who can be clustered for further analysis...
Or did you mean the individual aa properties and not the peptide on its whole?