A biomedical imaging system requires timely disease identification in order to avoid major risks for critical care patients. In most cases, collecting a biopsy sample from the patient, staining, histopathology imaging, and decision-making by an expert physician take a long time. If the image processing unit is part of the imaging system, a timely decision can be made. For efficient decision-making on CNN models, which are more expensive, a large amount of memory and a high-speed processor are required. If proper models that require less memory and high speed are identified, those models can be utilised in an imaging system for decision-making. This paper focuses on lung malignant tumour diagnosis using pertained deep networks to identify the network's capability to work on low-speed and low-performance devices. Ten pre-trained networks were selected based on module size, number of layers, and depth. Three networks were identified based on their high performance with moderate size, layers, and depth and accuracy of more than 90%.