I don't think I got your question. What are your random variable vectors? Are they x and y? If yes, then the covariance matrix would be a 2x2 matrix.
A 10x10 covariance is only possilble if:
You specify x as a 2x10 matrix (i.e. there are 10 random variable vectors, each with two instances) and compute cov(x).
You specify y as a 2x10 matrix (i.e. there are 10 random variable vectors, each with two instances) and compute cov(y).
Note that if you specify both x and y as above and copute cov(x,y), you'll still get a 2x2 matrix as the random variable vectors are now x and y.
In all of the cases, it doesn't make sense. The only sensible covariance I can think of is taking the covariance between the x coordinates of data sets x and y, and similarly the covariance of the y coordinates of data sets x and y.
%initialize x and y each with ten coordinates
x = randi(5,2,10);
y = randi(5,2,10);
%Three not very sensible covariance matrices:
%They are not sensible as the x and y coordinates of each
%data set is taken as an instance of the random variable
%even though they represent two different dimensions.
covXY = cov(x,y) %returns a 2x2 matrix
covX = cov(x) %returns a 10x10 matrix
covY = cov(y) %returns a 10x10 matrix
%Two sensible covariance matrices
covXY_x = cov(x(1,:),y(1,:)) %returns a 2x2 matrix
covXY_y = cov(x(2,:),y(2,:)) %returns a 2x2 matrix
