Even if it would be technically possible, it is not at all advisable. You are discounting the fact that MEAM is a manybody potential and thus you cannot simply subtract and add contributions on a per atom type basis. In the case of MEAM (or EAM) each substyle will only operate on a subset of atoms and thus also collect the contributions to the embedding term for only that subset, which is obviously incorrect.
Let me amend my last statement that the ultimate limitation is a limitation in pair style meam/spline that will prevent you from using it the way you intend to use it, even if you address all the errors and missing pieces in those few lines of input and your LAMMPS binary.
It doesn’t happen often that a limitation of a feature in a software is a good thing and prevents people from doing something that is bad.