After trimming your 3' adapters and filtering the reads for quality you map them back to reference genome --using a specialized aligner like SHORE or the ususal BWA, Bowtie etc.nThen you can filter them based on length to choose only miRNAs, compare these against miRBase to identify known miRNAs. For these known ones, you can compare expression across samples using standard diff. expression packages like DESeq or edgeR. For rest, use prediction programs to identify novel miRNA like -- mirDEEP.
There are also some online tools -- miRanalyzer and software packages -- miRcat and miRexpress for complete analysis and prediction steps.
Several webserver tools and software for miRNA expression profiling are discussed here -- http://www.springerlink.com/content/qg12476136718094/