That's the right direction, but I would also try to stay within the HISE user preset system so you get the intended loading prodecure (kill voices, load the preset on a background thread, then unsuspend the audio processing).
With the custom data model you can define the layout of the user preset file as you wish. Note that it will be converted from JSON to XML at some point though (which is a bit weird).
They also want to be able to export a preset, and include any sample content with the preset - ie; bundle the custom loaded .wav file, or the samplemap, with the preset. For personal sharing purposes.
I would detach this from the user preset system. A preset with sample mapping data and samples should be
The preset file
The JSON (or SFZ or whatever) mapping file
The audio samples
all in one folder.