Have a look at the FAQ section in the vignette. There is a section about pool of normals.
For coverage normalization, no, you cannot mix different capture kits. Use as many normals as you have. You can use technical replicates for coverage normalization, i.e. they do not need to be from different individuals. Less than 5 normals is not ideal, but can work if the data are of high quality.
For artifact removal it might make sense to run your pipeline on all the normal samples you have, even generated by different (but overlapping of course) assays. It's important though that the pipeline is in general the same, ideally also read lengths. Then merge all normals into a single multi sample VCF (e.g. with GATK3 CombineVariants) and follow the tutorial in the Quick vignette.