I would strongly suggest to stick to either a kit or an optimized and established protocol for standard applications such as 16S (or any NGS) rather than trying custom solutions unless absolutely necessary, e.g. from Illumina. The Illumina protocol uses the KAPA HotStart, and this is a high fidelity one with 3->5' exonuclease activity, so it leaves no overhangs.

I can say by personal experience that it is often tempting to try out something new rather than "boring" and established kits, but if the prep then fails you essentially have no clue what the problem was and you have to a) either use that boring kit anyway or 2) start optimizing and debugging your entire protocol. Make life easy, use a kit, and spend the saved time on the science, and be it just to read more papers, or maybe learning a new programming language (not Perl please) ;-)