In this work, we investigate a Recurrent Spiking Convolution Neural Network (RSCNN) with Adaptive Spiking Neurons (ASNs) that is enabled with adaptive threshold, spiking frequency, and surrogate gradient (ASG) slope, evolved from neuro-evolutionary scheme adopted spiking neural architecturesearch (NAS), to classify three primary emotional states: angry, happy and sad. The proposed neuro-evolved NAS search in the form of Adaptive Deep Evolutionary SPIking NEtwork (DESPINE) configuration is designed for edge deployment, seeking an energy efficiency of 9.4x and yet attain a near state-of the-art (SOTA) accuracy of 85.36% (SOTA is 86.3% on a GPU, consuming ~35 W on a P100 GPU and ~2.8 W on an edge-GPU ) for the model trained on FER-13 dataset with 64 timesteps. The EDP gain of 19.42x was achieved over the SOTA methods. The adaptive neuron model scheme with the NAS run extracted SNN model offers 3.13x energy savings over the corresponding NAS evolved SNN model comprising of non-adaptive neuron model. The proposed DESPINE framework was evaluated for MNIST, CIFAR-10 and Neuromorphic-MNIST (NMNIST) datasets towards achieving SOTA accuracy, in a view to showcase the suitability of the approach for diverse datasets. The low power costs of the SNN is beneficial to healthcare bound diagnostic solutions, where the device and the process are continuously running in the background for a long duration. All the scripts and extracted SNN model files are made freely available for further usage to the researchers and designers' community.