Hi,
counting reads per exons is not trivial.
Typically, exons belong to many overlapping transcripts, so the same exon repeats multiple times in the GTF file.
Also, you need to deal with the overlapping exons.
If you can create a set of non-overlapping "super-exons", you can do the counting on them instead of genes.