One can't possibly get rid of all the biases in the data during RNA seq analysis, but depending on what kind of interpretations you desire, you can choose the experimental workflow and the normalization procedures to make the most out of your data(e.g between-sample normalization adjusts for effects related to distributional differences in read counts between samples, e.g. sequencing depth, library preparation while within-sample normalization adjusts for gene-specific and sample effects e.g. related to gene length and GC content). There is no sense in double normalizing data. You should decide what normalization procedure will be best suited for your needs, and if you can't decide upon it, I'd recommend using scone framework(Cole et al., 2019, Cell Systems 8, 315–328)- it is used for evaluating and ranking the different normalization procedures using some performance metrics considering different aspects of a desired normalization outcome.