That said, to my eyes the primary problem with how we teach calculus is that we introduce too many big concepts at the same time. The key picture that people should come away with is the tangent line. But when you say that the derivative is a function defined at each point as the slope of the tangent line at that point, most people experience brain freeze.
Instead I would like to see students get to the point where they can calculate tangent lines. Then stop. Work out lots of tangent lines for lots of functions. Introduce them to how the tangent line can be used to estimate error bars. (Typical word problem, if a rock take 5 +- 0.5s to fall off a cliff, how tall is the cliff?) And then only after the picture and mechanics of the tangent line is well established should we move on to defining the derivative.
This approach, if done right, would be rigorous but informal. It should avoid the problem of causing brain freeze by presenting too many layers of generality at once. And it would really cement the key image behind half of calculus.
I think the general point is true about the brain freeze and too many big topics -- it's way too easy to just throw a definition out there, say "this represents slope which is the rate of change" and move on to the next idea.
I think the most important thing in the beginning is to learn to think conceptually about calculus, about how to break a shape into a simpler model and solve it (not just manipulate models that are given to you). Later on (in analysis, which I've never formally taken :) ) the details about the transfer principle, epsilon-delta proofs, etc. can emerge, but realistically only a very, very small fraction of students need that. That fraction can be increased if they are able to use calculus regularly (with infinitesimals, say) and then wonder the details of how it all works.
I like to ask: Would it be such a bad thing if everyone understood calculus as Newton did, at a baseline level? Many people taking calculus leave with no understanding whatsoever, but the same isn't said for geometry or even algebra (you can make references to parallel lines, equations and feel that people will 'get it', but mention a derivative or limit to someone 10 years after calc class and you'll get a blank stare). I advocate having rough mental models that are successively refined over time (vs. presenting the most rigorous version first).