Here's a report which talks in great detail about the APRS protocol.
http://digitalcommons.calpoly.edu/cgi/viewcontent.cgi?article=2449&context=theses
That looks like a really useful document, one that I could've done with fifteen years ago. Documentation around all amateur packet radio facets has traditionally been steeped in folklore and voodoo: only a very few ever really understood it to this level of detail.
I'd also recommend that the OP look up FX.25, which is a backwards compatible FEC implementation of AX.25. Possibly the biggest drawback of AX.25 is the lack of FEC. Another is the modulation and coding scheme, but changing that would preclude backwards compatibility.
I haven't read it all, but as an aside I did find it somewhat amusing that the author fights to map the stack to the ISO OSI straitjacket model: it looks like academia continues to peddle this massively over-egged and largely unnecessary abstract model made up of word soup that you learn by rote. I remember it being preached to me back in the early 80s at university, and when I asked for a practical example of it so I could better understand the abstraction, it was almost a tumbleweed moment. Since then ethernet and then TCP/IP have become unbiquitous and de facto, but both at the time and since then, there have been many other stacks that people feel the need to shoehorn into the OSI model, possibly the worst being the multiple protocols inside IBM's SNA architecture which were even more ill-fitting.