Mohammad - I don't think there is any problem using a rank transformation and one-way ANOVA. There are a number of papers on this - including the classic work by Conover and Iman. This is probably one of the easiest ways to run post hocs for KW tests in SPSS.

There are problems with factorial models because conceptually an interaction on the ranked data does not imply an interaction on the original scale or vice versa.

I have done as you suggested in the past-- used Mann Whitneys for pairwise comparisons after Kruskall Wallis tests.  Adjusting for Type I errors is appropriate and conservative.

You can also use pairwise wilcoxon test using Bonferroni adjustment. In case you use R, the command is:

pairwise.wilcox.test(variable,group,p.adj="bonferroni")

AK Singh, thanks for showing the pairwise procedure with Wilcox.

On pairwise.wilcox.test, the default method, Holm, is generally preferred by methodologists to Bonferroni (in statistical terms it dominants it).

You are absolutely correct, "p.adj=holm" is considered to be superior, but I don't know which test is considered 1st, which 2nd etc. so I always recomment Bonferroni. 

I would actually reccomend the Hochberg procedure (which is in effect a variant of the Holm test). It is more powerful, but less general. However, for pairwise testing (where the test statistics are necessarily positively correlated) the Hochberg procedure's assumptions are met.

The Holm procedure is still useful for controlling multiple tests where the test statistics are not positively correlated.