The work presents a hardware aware modeling and learning rule for mixed-signal Spiking Neural Network (SNN). The Python-based models are compared with transistor simulation in 65nm CMOS technology. The general approach to training and inferring with mixed-signal SNN is to learn the weights offline and then deploy them on neuromorphic hardware. However, in the case of analog or mixed-signal hardware, the circuits employed for computation are non-linear and have significant variability. This work is an initial proof-of-concept towards fully modeling this non-linearity and learning with these circuits. The study specifically employs Floating-Gate (FG) transistor-based synapses for storing the weights of the synapses and adaptive Leaky-Integrate and Fire circuit for modeling the neurons in the SNN. Additionally, the study develops a localized gradient-based algorithm to learn the FG voltage of the analog synapses SNN. This approach enables integrating the non-linearity of analog synapses into the learning framework. Using these models and learning algorithm the work shows system-level classification using multiple layers, neurons and time-steps.