Previously on Law & Order: Re^3: Design Question - I've gone object mad!.

You seem to be saying the tools out there are beyond your ability to use so you'll write new tools. There is some cognitive dissonance going on in there. One small example: you mention detecting primary keys being auto increment. What about multiple column keys? Already, out of the gate, the problem is much harder than it was a second ago and not because the requirements changed but because your awareness of the requirements changed.

You could probably cobble together something useful for you own particular problem but it certainly won't be as good as what's out there already. If your goal is to learn more about code development and design, go for it. If your goal is to get work done and sharpen your employability, get thee to a CPAN node.

Some of the available tools are indeed quite difficult to learn, like DBIx::Class but that's only because they do so much and have considered, tested, vetted so much.

As you mentioned, SQL::Abstract is nice and its a good gateway to DBIC if you end up wanting more. It can't do auto-discovery like DBIC but it's certainly easier to master.