The kmer length is not too much of a factor given it is shorter than the read length in my experience. I did a small test a while back, see Salmon Quantification for RNA-seq Read Pairs with Different Lengths and the influence of different lengths is minor. It all comes down to the libary quality. In sloppy language: If the sample is crap, so will be the mapping/quantification. The important factor is that the library is good quality and read length is sufficient. Therefore, decide for a kmer length, e.g. 25 and use it for your analysis.