You'll need to add
--enable-experimental --enable-spatial-svc to your configure to get examples/vp9_spatial_svc_
encoder to compile.
Prediction from different sized reference frames is an integral part of VP9 and as such the mechanism is implemented in libvpx. It is up to the application to decide how to make use of this functionality.
vpxenc allows the user to scale *every* frame in the sequence to the same size by specifying the flags --resize-allowed=1 --resize-width=XXX --resize-height=YYY.
However, vpxenc does not allow variable frame scaling at all. If you look at examples/vp9_spatial_svc_encoder.c that provides an example of how to handle dynamic scaling in the context of a spatial scalable application.
These are only two specific implementations and other use-cases are very possible. Examination of examples/vp9_spatial_svc_encoder.c should help you to understand how to use lkibvpx to achieve your own desired solution.
Is this the information you needed?
Adrian