In this era of modern medicine diagnostic practices, the images of brain tumors using the MRI machine is a very important practice, as if detected early and correctly, it may be a lot for the treatment. In order to classify four categories of brain tumors, namely glioma, meningioma, pituitary tumor and no tumor, this research adopts the idea of multi-model Deep Convolution Neural Network (D-CNN) that is DenseNet121, DenseNet201, NeuroFusionNet (EfficientNetB0+MobileNetV2), Ensemble model and ResNet152V2 resnet respectively. To make the model easier to generalize, a dataset of MRI images is publicly available on Kaggle, and this dataset has preprocessed in the form of augmentation, normalization, and scaling of pictures. Explainable AI (XAI) employing Grad-CAM was included to determine the discriminative areas that affect the predictions. All the models were meticulously trained and tested in terms of conventional performance metrics. ResNet152V2 was able to perform feature extraction and were resilient while classifying multi-class tumors, rendering the best accuracy rate that was seen among all the models, which was equivalent to 98.93%. Furthermore, the good and constant results acquired by the Ensemble and NeuroFusionNet had been demonstrated the utility model fusion. Grad-CAM heatmaps made the outcomes of automated diagnosis more interpretable and trustworthy since it is clinically important visual explanation. All things considered, the proposed approach is a good and reliable structure that can be used to help radiologists diagnose brain tumors early on.