Plant disease affects crop productivity and yield. Several machine learning and deep learning architectures have been used for plant disease classification for the last decade. Tomato and Potatoes are being grown in many parts of the world. This research study presents various pruned deep CNN models for potato and tomato disease classification where different CNN architectures were pruned, their core structure was retained, and parameter size was reduced to approximate 1.5 M. Rather than focusing solely on accuracy, this study also considers reproduction and deployment. Images of tomato and potato plants covering 13 diseases, including healthy, were taken from the PlantVillage dataset. Six deep CNN models were considered, pruned and concatenated, out of which EfficientNetB0 have a superior performance score of 99.90%, with a parameter size of 1.5 M and 0.826G computing operations (FLOPs), outperforming the remaining five pruned CNN models. The developed lightweight model can be easily deployed in mobile and portable devices with limited resources and computational power, to assist farmers in identifying potato and tomato diseases.