I'm analysing RNA-Seq data of patient samples taken pre and post chemo. Should these groups be normalized separately or together?

Normalization is always conducted for all samples together. Anything else would defeat the purpose of normalization, which is to put all the samples on a comparable scale.