Supposing you have time- derivatives then the highest derivative order "n" is the dimension of the state vector ( taht is the order of the differential equation describing the system´s dynamics) . The state- space description can be as usual : the first state component is the solution ( or output) and the succesive derivatives uptlil order (n-1) are the remaining ones. If the dynamics is linear and the input is nonlinear with a separate nonlinearity ( for instance , a saturation) , this would be simpler. Simply describe the input to state and output mappings in the standard linerar state-space fashion and then incorporate the function descibing the nonlinearity to a feedback information mapping of the form : output or state to input.
There is no general strategy for the nonlinear case. It depends on each system.
I am adding other two papers. a) saturation with a particualr modelling of it, and b) nonlinearity being a nonlinear perturbation term in the state space. If you have a differential equation of order n then the state space has a dimension equal to n. If the nonlinear is a separate one outside the controlled plant then the control can take into account of it.