There are probably other ways, but here's a way (if I understand the question correctly).
A = rand(6,3,6);
dnn = 5; % do not normalize this row
Adn = A(dnn,:,:); % save this row
A = A./sum(A,2); % normalize
A(dnn,:,:) = Adn; % replace the row
sum(A,2) % show that the row sums are 1 except where specified
That will normalize all rows except the specified row. I should point out that while this method discards the normalized results for one row, it's faster than indexing
