A technique for recognising and labeling malignant brain tissues according to the types of tumours present is known as tumour classification. Magnetic resonance imaging (MRI) can be used in clinical settings to both diagnose and treat gliomas. For clinical diagnosis and treatment planning, the ability to correctly diagnose a brain tumour from MRI images is essential. Manual classification, however, is not feasible in a timely manner due to the enormous volume of data produced by MRI. For classification and segmentation, it is required to employ automated algorithms. However, the numerous spatial and anatomical differences present in brain tumours make MRI image segmentation challenging. We have created a unique CNN architecture for classifying three different types of brain cancers. The new network was demonstrated to be more straightforward than earlier networks using MRI images with contrast-enhanced T1 pictures. Two 10-fold cross-validation techniques, two datasets, and an evaluation of the network's performance were used. A piece of upgraded picture information is used to assess the transferability of the network as part of the subject-cross-validation process. When used for record-wise cross-validation, this method of tenfold cross-validation ground set has an accuracy rate of 92.65 percent. Radiologists who operate in the ground of medical diagnostics may find the newly proposed CNN architecture to be a helpful decision-support tool due to its new transferability capability and speedy execution..