Ive worked on a project like this. Not the programmable logic but on the software side. The problem was simulair to this. The approach we took was like this.
To recreate the exact functionality of exsisting hardware lab proto's were buidt that could interface with the original hardware. For instance; you pop out the device you want to replace. Build a protoboard that connects to the original interface (or directly to the PCB) The protoboard contains glue circuitry and your programmable device. The glue is to cope with electrical incompatibilities between the new decive and the current interface.
You program your new design in your programmable device. When your design is functionally the same and all works as before you advance to the next device. After all devices are redesigned into seperate parts you can take the next step.
Melting your design into one fpga. At this point you can estimate how much of your design will fit into your target device. You could decide to split things in order to be more expandable.
Ofcourse this will require some redesigning because interfaces will change again. At least your functional blocks are correct. You verified them using the existing hardware.
When you have melted the blocks into your target devices, redesigned the PCB (with expandability in mind, and got it all working again you can add nice features to the design.
But I would first of all keep it simple.
A collegue of mine made a drop in replacement for the 6510 CPU this way just for fun. He has been expanding that design to a more modern architecture (he said he got to 68040 comparable performance) but the project is in the ice cabinet now since he has become father recently.
Still, I think these ventures are technically possible, but as a one man project it is probably too much work.
It also requires some cash. When you run into trouble (and you will) you need pretty expensive equipment to measure all the signals you are interested in.
But it is always nice to think about these projects.
