Those best-practice guidelines are for variant calling, not gene expression. BSQR would be useless, and indel realignment might even cause problems for alternatively spliced transcripts b/c it's designed to resolve alignment discrepancies at indel loci.

The GATK pipeline for RNA turns reads that are split over an intron into two (or more) individual reads. This is done specifically so you can run BQSR and IndelRealigner on RNA-Seq data. However, as Harold points out, this is far from an established protocol. It could, depending on the data, do more harm than good. As H.Hasani mentions, the only real way to know for sure is try both and find out. I know thats kind of easier-said-than-done... but you're doing cutting-edge research here, so it is unlikely things will go as smoothly as we'd all like. Good luck :)