I made a pipeline adapted to small RNA-seq with multimapped reads. It is based on bowtie2, allowing reads to map at up to 10 locations (option -k 10), but keeping only best-scoring ex-aequo. If I remember well, I divided the contribution of a given read to the total counts by the number of such ex-aequo locations. The counting is done in a custom python script based on pysam.

The pipeline is here: https://bitbucket.org/blaiseli/pirna-pipeline

It is for Drosophila and we were not interested in tRNAs, so the counting of tRNAs is only global, not at the gene level. We were more interested in piRNAs. We counted them in a second step by re-mapping the reads identified as piRNA candidates to piRNA clusters and transposable elements. Besides, I was then unaware of lots of tools, formats and ways of doing common in bioinformatics, so my approach may sound complicated. It would probably be simpler to write your own counting script, but feel free to fork, adapt, use ideas...