Most of the analysis is similar (e.g.: QC, count generation, differential expression etc...). Due to differences between the interments, you will typically see reads with lower Phred scores from the QC, but that's expected.

The primary difference is in mapping due to homopolymer reads. Check out this post for some discussion about this and suggestions on parameters. Additionally, you may consider using the "Two-Step Alignment" approach briefly discussed here for Ion Torrent data, but I suggest you search the web further for additional details on this method. Admittedly, the information about this is a bit less than the Illumina systems, but it's there if you dig in. In short this method suggests to use a spice-aware aligner to map the reads and then take the unmapped reads and use an aligner like Bowtie2 to complement the mapping (the poor mapping from the splice-aware aligner is usually due to the homopolymer reads). And then of course you would need to combine the BAM files and sort with samtools.

I would suggest to explore this further and benchmark some strategies and see what works best. There may be better approaches to deal with Ion Torrent data now (and perhaps quasi-mapping could do a good job? I have not tested this yet).