You can implement a string encryption method by your own. Using Java you need 3 or 4 lines of code only. Unfortuantely the export laws do not allow to publish the code on a server located in the USA, because evil countries could use it to encrypt their evil data also.
But if you ask your favourite search engine for "MATLAB AES encryption", your are still successful, I think. If you are a citizen of an evil country, please do not ask a search engine. (Any additional details of my personal opinion about the US export restrictions would be off-topic.)
Then you have the standard problem, that the encryption key must be stored reliably also. Never store the password in clear text, even not in P-coded files, otherwise the decryption gets near to trivial and a dull ROT13 method would offer the same level of security.
Another problem appears, when you need to process MAT-files: You can write the MAT file in clear text, encrypt it and ship it to the users. But then you cannot load the file directly. You have to decrypt the file at first and write it to the harddisk before load can import the data. But then you need a very sophisticated method to remove the clear-text data from the disk afterwards. I'm convinced, that you cannot reach a 95% security level by this way.
You find tools to "serialize" objects in the FEX (use this term to search there). This means, that e.g. a struct can be converted to a byte stream, which can be encrypted easily. Then the decryption can happen in the memory as well as the "deserialization". In consequence the hard disk remains clean and the clear text data can only be found using a memory dump. When the password is kept securely, this means a security level of 99% (in arbitrary units...).