Brain tumor is one of the most dangerous diseases, which is very hard to diagnose due to its rare symptoms. Diagnosing disease at right time helps to give proper treatment and could extend patient survival period. Histopathology images of brain tumor are taken from the Cancer Genome Atlas (TCGA). Large numbers of tissues have to be analyzed to diagnose disease efficiently, which produces time consuming problem. In this model CNN architecture like InceptionV3 and InceptionResNetV2 are adapted to solve binary and multi-class issues in brain tumor histology images using transfer learning. InceptionV3 is used to extract features and for fine-tuning, InceptionResNetV2 is used for feature extraction. Framed autoencoder network to transform the extracted features to low dimension space and to do clustering analysis on image. Proposed autoencoder produce better clustering result than features extracted by InceptionResNetV2.