Brain tumor is one of the diseases with less survival chances. There are different types of tumor found in human brain mainly classified as cancerous tumor or non cancerous tumor. Further these tumors can be distinguished on the basis of their type and location such as pituitary tumor, glioma tumor or meningioma tumor. Magnetic resonance images (MRI) is an imaging technique which is widely used to study brain tumors because of its various advantages over other imaging methods. As there are hundreds of brain slices in MRI images and to study those manually is a tedious task and can be a very time consuming process. To facilitate the medical experts, this research work proposed an automated brain tumor classification and segmentation process. The study focuses on two major tasks of brain classification and brain segmentation. An EfficientNetB7 neural network is used to classify brain tumors from the MRI images. The overall classification accuracy of the model achieved was 98.485 percent. Which is better than the other existing models such as VGG-16, CNN, MobileNet, ResNet152v1 and some recent works. For segmentation of brain images consisting tumor an effective non linear curve fitting based method is proposed using levenberg-Marquardt algorithm. The proposed method showed better results than the existing state of art methods.