high precision photo diode
I can't follow your view on "hand created" versus built-in models. Both are empirical to some extent.
On the other hand, for all device properties, that are not represented in SPICE standard models, equivalent circuits or analog behaviour descriptions are the most simple ways. Of course, you can also add user model code, if you have a version, e.g. original Berkley SPICE, that has the option. But this would be interesting rather to increase speed than achieve better accuracy. As far as I experienced, anything can be modelled as analog behaviour.
The real problem is to get correct model parameters for your device. If you want to measure it's behaviour first and extract a model, why not build your intended PIN diode circuit, make some measurements and you're done. What's the particular purpose of simulation in this case?