Could a code be developed where one uses intentional errors say in english, as a text to encode? For example someone might have a message 'Agent X must report to station 5'. This could be distorted to 'A Gentex missed repaired 2 stationS.' The distorted message then encoded. If the distortions are recognizable to the deciferer this might throw off any attempted interceptions. Also aren't decoding methods usually based on the 'regular' use of english (if english is being used) ? I've never heard of anybody trying to use intentional mistakes in a code. Is this idea useful?
The question has been answered but let me add some historical examples.
Enigma operators often made spelling mistakes; it didn't help their cause. Navajo code talkers on the other hand were a good measure even if the cipher was broken.
What you are describing is very close to a secret language. A certain terrorist organization operating in Greece in the '90s used such a language, for example they'd use the word "basket" instead of "bomb" (you won't find that in the Wiki article, I'm reporting from memory). I'm sure some cursory digging would reveal other organizations which have done the same thing - terrorist and otherwise.
So it's essentially a word substitution cipher with some spelling noise. A parser looking for natural languages will flag it. A dictionary parser will underline it and a codebreaker will scratch his head for a bit and eventually figure it out.
That is, assuming you're operating some kind of manual cipher and haven't read a book on cryptography (like the terrorists above). Fortunately we have much better alternatives. Regarding your comment:
The simple answer is no. A good cipher is language-agnostic (actually, pattern-agnostic). This means that a well-chosen plaintext (or a number of them) cannot undermine the security of the protocol (see CPA). The Solitaire cipher for example is meant to be used with a deck of cards, where each card corresponds to a letter of the Latin alphabet. But nowhere does it assume that it's shuffling a natural language, much less English. Each card has a certain numeric value; how the operator perceives these values is not the cipher's concern.
The short answer is that such an encoding would not help with encrypting the content.
A good, modern encryption scheme should at least be secure if the attacker can choose the plaintext, let alone know what the plaintext is. Moreover, most of them are secure if the attacker can even ask for the decryption of ciphertexts of his choice (other than the actual challenge message they're trying to decrypt).
I think the concept you are describing is steganography, with deliberate typos/mistakes in a text document hiding the fact that there is any data to be found.
This technically does not encrypt anything (it's a form of encoding), but could be used to store/send an encrypted message in a way that might avoid rousing suspicion. See http://en.wikipedia.org/wiki/Steganography
To stand a chance of being decoded, your deliberate mistakes would need to be simple to detect by a decoder. A really simple scheme for encoding might take every fifth word (defined as any group of letters with spaces or punctuation on left and right) and treat it as 1 bit - it would be a 0 if summing the ASCII values of the letters was even, and 1 if it was odd. If the word already in the document was wrong under that scheme, then make some randomised edit to get the bit value correct for the message you wanted to store.