As the world becoming so much internet de-pendent and near about all the communications are done via internet, so the security of the communicating data is to be enhanced accordingly. For these purpose many encryption-decryption algorithms are available and many neural network based keys are also available which is used in these algorithms. Neural Network is a technique which is designed to work like a human brain. It has the ability to perform complex calculations with ease. To generate a secret key using neural networks many techniques are available like Tree Parity Machine (TPM) and many others. In TPM there are some flaws like less randomness, less time efficient. There are already three rules available i.e. Hebbian Rule, Anti Hebbian Rule and Random Walk, with same problems. So to overcome these issues, we propose a new approach based on the same concept(TPM, as Tree-structured Neural Network’s execution time is comparatively less than that of the other Neural Networks) which generate random and time-efficient secret key.

Sharing of data is always a crucial job in the world of data technology, where data are perceived as useful resources for generating significant information for taking notable decisions or for doing interpretation. On the basis of impact factor associated with the information, sharing and storage of data demands security. Sharing of large extent of textual sensitive data contained in a file needs a way of hidings their direct decipherment. Due to the size related restriction associated with steganography we always promote cryptography for sharing large amount of sensitive data. In this paper we have proposed an encryption technique with time complexity of O(1) that exploits the dynamic node fusion property of graph theory. We are applying the rule of cosmos which reveals that the entire cosmos relies on the conception of data hiding. Each time billions of particles merges to form a new structure and again splits, gets distributed and again follows the same process of hiding and disclosing the hidden truth in a cyclic manner. Following this logic we have proposed a dynamic layered encryption technique that will be completely able to resist the illicit actions of intruders with low computational efforts as well as it reduces the network load on packet transmission. In future with the successive increase in the processing power and requirements we can easily intensify the capacity of the proposed technique.