I have (poly-A) mRNA-seq data. I want to have RPKM/FPKM values for each gene and I want to provide a cufflink with a gtf file with only exons (or should the transcript information be also included?) annotation in order to get rid of the reads that can fall into intronic/intergenic region. What risk can into with this approach?
I believe there is some confusion between read mapping (that is usually done on the whole genome), and the RPKM/FPKM calculation that is always made on specific features, such as exons.
In your case, Cufflinks take as input reads that are already mapped, so you don't "get rid" of reads that fall into intergenic regions, you'll just ignore them if they don't fall in regions of interests. Now, to answer your question, I think it is fine to input only exons if you really want to. But why not give cufflinks annotated transcript too ? The more information, the better. If you are worried about de novo transcript discovery, cufflinks can do it even if you give him annotated transcripts so its fine !