You can use symbolic math tool box of matlab, e.g. you can do
>> b = sym('321.097808043632864e432');
>> a = sym('123.87698272893e445');
>> a*b
ans =
3.977662762131838872558175155552e881
Obviously you can mix this with any other matlab module, however note that these operations are very expensive. 128-bit real (up to 10^4932) is another option but I don't think MATLAB supports that. Fortran supports quadruple percision and you can write a Fortran subroutine and connect from Matlab to that piece of code using MEX files. The only remaining issue would be to send and get the 128bit real in and out of your Fortran module which I am not sure how that can be done.
It depends on the volume of data that you have to work with.
But, if you cannot escape from Matlab you can always rescale your data by dividing them with the greatest entry in order to rescale them to [-1,1] range.
The best option is to represent the numbers in polynomial form and write a small program in matlab to do the neccessary mathematical operations. This can be done faster if you concept of Representa arithmetic for addition, subtaction and multiplication of any two arbitrarily large numbers.